\bigskip         \subcaption*{\normalsize Panel B: Logit}         \vspace{-0.5ex}         \begin{tabularx}{\hsize}{lCCCCCCCC}         \midrule
                    &\multicolumn{1}{c}{(1)}         &\multicolumn{1}{c}{(2)}         &\multicolumn{1}{c}{(3)}         &\multicolumn{1}{c}{(4)}         &\multicolumn{1}{c}{(5)}         &\multicolumn{1}{c}{(6)}         &\multicolumn{1}{c}{(7)}         &\multicolumn{1}{c}{(8)}         \\
\midrule
HEE                 &       0.610         &      -0.172         &       0.290         &      -0.511         &      -0.067         &      -0.054         &       0.300         &       0.913\sym{***}\\
                    &     (0.446)         &     (0.408)         &     (0.184)         &     (0.612)         &     (0.477)         &     (0.362)         &     (0.196)         &     (0.215)         \\
HEEC                &       0.466         &      -0.626         &      -0.086         &      -0.477         &       0.541         &       0.462         &       0.001         &       0.893\sym{***}\\
                    &     (0.458)         &     (0.471)         &     (0.174)         &     (0.612)         &     (0.578)         &     (0.313)         &     (0.204)         &     (0.213)         \\
\midrule
Observations        &         839         &         839         &         839         &         839         &         839         &         839         &         839         &         839         \\
\midrule
\end{tabularx}
\captionsetup{justification=justified,
width=\hsize}
\caption*{\footnotesize
\textit{Notes:}
This
table
examines
the
robustness
of
the
results
presented
in
Table
\ref{table-endline-behavior}
to
using
probit
or
logit
regression.
The
specifications
do
not
include
randomization
strata
fixed
effects
because
success
or
failure
is
perfectly
predicted
within
some
cells.
***$\,p
<
0.01$,
**$\,p
<
0.05$,
*$\,p<0.10$.
}
\end{table}
